A novel bidirectional mechanism based on time series model for wind power forecasting

A novel bidirectional mechanism and a backward forecasting model based on extreme learning machine (ELM) are proposed to address the issue of ultra-short term wind power time series forecasting. The backward forecasting model consists of a backward ELM network and an optimization algorithm. The reverse time series is generated to train backward ELM, assuming that the value to be forecasted is already known whereas one of the previous measurements is treated as unknown. In the framework of bidirectional mechanism, the forward forecast of a standard ELM network is incorporated as the initial value of optimization algorithm, by which error between the backward ELM output and the previous measurement is minimized for backward forecasting. Then the difference between forward and backward forecasting results is used as a criterion to develop the methods to correct forward forecast. If the difference exceeds a predefined threshold, the final forecast equals to the average of forward forecast and latest measurement. Otherwise the forward forecast keeps as the final forecast. The proposed models are applied to forecast wind farm production in six time horizons: 1-6 h. A comprehensive error analysis is carried out to compare the performance with other approaches. Results show that forecast improvement is observed based on the proposed bidirectional model. Some further considerations on improving wind power short term forecasting accuracy by use of bidirectional mechanism are discussed as well.